Re: [PATCH v3 2/2] builtin/pack-objects.c: introduce `pack.recentObjectsHook`

> I think this description suffers a bit from being adapted from the
> original patch which was targeting cruft packs. It's not clear to me
> what "the caller" means here. And really, I think this is getting into
> the details before giving an overview and motivation.
>
> I'd expect something the rationale to be something like:

> One option I don't see here is: update the mtime on the objects you want
> to salvage.
>
> Why would we want this patch instead of just having the caller update
> the mtimes of objects (or in a cruft-pack world, call a command that
> rewrites the .mtimes file with new values)?
>
> I can think of some possible arguments against it (you might want to
> retain the old mtimes, or you might find it a hassle to have to
> continually update them before gc kills them). But I think the commit
> message should probably make those arguments.

> > We then add those as tips to another reachability traversal (along with
> > any recent objects, if pruning), marking every object along the way
> > (either adding it to the cruft pack, or writing it out as a loose
> > object).
>
> I didn't understand this "if pruning" comment. If we are not pruning at
> all, wouldn't we skip the extra traversal entirely, since we know we are
> saving everything?

> > @@ -126,8 +198,14 @@ static int want_recent_object(struct recent_data *data,
> >  			      const struct object_id *oid)
> >  {
> >  	if (data->ignore_in_core_kept_packs &&
> > -	    has_object_kept_pack(oid, IN_CORE_KEEP_PACKS))
> > +	    has_object_kept_pack(oid, IN_CORE_KEEP_PACKS)) {
> > +		if (!data->extra_recent_oids_loaded)
> > +			load_pack_recent_objects(data);
> > +		if (oidset_contains(&data->extra_recent_oids, oid))
> > +			return 1;
> > +
> >  		return 0;
> > +	}
> >  	return 1;
> >  }
>
> This hunk I'm less sure about. The purpose of this function is that the
> caller has told us about some packs which are "special", and we avoid
> adding their objects to the traversal.
>
> This kicks in for cruft packs, when the git-repack caller says "I just
> made pack xyz.pack; do not bother saving anything in it to the cruft
> pack, since xyz.pack is here to stay". So if a hook says "you should
> keep object X", why would we want to override that check? It is already
> a reachable object that has been packed into xyz.pack, so we know there
> is no point in even considering its recency.

> > --- a/t/t5329-pack-objects-cruft.sh
> > +++ b/t/t5329-pack-objects-cruft.sh
> > @@ -739,4 +739,175 @@ test_expect_success 'cruft objects are freshend via loose' '
> >  	)
> >  '
> >
> > +test_expect_success 'additional cruft tips may be specified via pack.extraCruftTips' '
>
> This title (and others below) seems out of date. :)

> > diff --git a/t/t7701-repack-unpack-unreachable.sh b/t/t7701-repack-unpack-unreachable.sh
> > index ebb267855f..d2eea6e754 100755
> > --- a/t/t7701-repack-unpack-unreachable.sh
> > +++ b/t/t7701-repack-unpack-unreachable.sh
> > @@ -113,6 +113,28 @@ test_expect_success 'do not bother loosening old objects' '
> >  	test_must_fail git cat-file -p $obj2
> >  '
> >
> > +test_expect_success 'extra recent tips are kept regardless of age' '
> > +	obj1=$(echo one | git hash-object -w --stdin) &&
> > +	obj2=$(echo two | git hash-object -w --stdin) &&
> > +	pack1=$(echo $obj1 | git pack-objects .git/objects/pack/pack) &&
> > +	pack2=$(echo $obj2 | git pack-objects .git/objects/pack/pack) &&
> > +	git prune-packed &&
> > +
> > +	git cat-file -p $obj1 &&
> > +	git cat-file -p $obj2 &&
> > +
> > +	write_script extra-tips <<-EOF &&
> > +	echo $obj2
> > +	EOF
> > +	git config pack.recentObjectsHook ./extra-tips &&
> > +
> > +	test-tool chmtime =-86400 .git/objects/pack/pack-$pack2.pack &&
> > +	git repack -A -d --unpack-unreachable=1.hour.ago &&
> > +
> > +	git cat-file -p $obj1 &&
> > +	git cat-file -p $obj2
> > +'
>
> And this is the new test in this iteration covering the "repack -A"
> case.
>
> It is checking that $obj2, which our hook mentions, is saved. It also
> checks that $obj1 is saved because it is still recent. But there are two
> other possibly interesting cases:
>
>   - an object that is too old and is _not_ saved. It seems useful to
>     confirm that the new patch does not simply break the ability to drop
>     objects. ;)
>
>   - an object that is reachable from $obj2 is also saved. From a
>     white-box perspective this is less interesting, because we should
>     already test elsewhere that this works for recent objects, and we
>     know the new feature is implemented by faking recency. But it might
>     be worth it for completeness, and because it's easy to do (making
>     $obj2 a tag pointing to a blob should work).